On Wed, Feb 09, 2022 at 11:13:15AM IST, Alexei Starovoitov wrote:
> From: Alexei Starovoitov <ast@xxxxxxxxxx>
>
> The main change is a move of the single line
> #include "iterators.lskel.h"
> from iterators/iterators.c to bpf_preload_kern.c.
> Which means that generated light skeleton can be used from user space or
> user mode driver like iterators.c or from the kernel module or the kernel itself.
> The direct use of light skeleton from the kernel module simplifies the code,
> since UMD is no longer necessary. The libbpf.a required user space and UMD. The
> CO-RE in the kernel and generated "loader bpf program" used by the light
> skeleton are capable to perform complex loading operations traditionally
> provided by libbpf. In addition UMD approach was launching UMD process
> every time bpffs has to be mounted. With light skeleton in the kernel
> the bpf_preload kernel module loads bpf iterators once and pins them
> multiple times into different bpffs mounts.
>
> Acked-by: Yonghong Song <yhs@xxxxxx>
> Acked-by: Andrii Nakryiko <andrii@xxxxxxxxxx>
> Signed-off-by: Alexei Starovoitov <ast@xxxxxxxxxx>
> ---
> kernel/bpf/inode.c | 39 ++----
> kernel/bpf/preload/Kconfig | 7 +-
> kernel/bpf/preload/Makefile | 14 +--
> kernel/bpf/preload/bpf_preload.h | 8 +-
> kernel/bpf/preload/bpf_preload_kern.c | 119 ++++++++----------
> kernel/bpf/preload/bpf_preload_umd_blob.S | 7 --
> .../preload/iterators/bpf_preload_common.h | 13 --
> kernel/bpf/preload/iterators/iterators.c | 108 ----------------
> kernel/bpf/syscall.c | 2 +
> 9 files changed, 70 insertions(+), 247 deletions(-)
> delete mode 100644 kernel/bpf/preload/bpf_preload_umd_blob.S
> delete mode 100644 kernel/bpf/preload/iterators/bpf_preload_common.h
> delete mode 100644 kernel/bpf/preload/iterators/iterators.c
>
> [...]
>
> -static int __init load_umd(void)
> +static int __init load(void)
> {
> int err;
>
> - err = umd_load_blob(&umd_ops.info, &bpf_preload_umd_start,
> - &bpf_preload_umd_end - &bpf_preload_umd_start);
> + err = load_skel();
> if (err)
> return err;
> - bpf_preload_ops = &umd_ops;
> + bpf_preload_ops = &ops;
> return err;
> }
>
> -static void __exit fini_umd(void)
> +static void __exit fini(void)
> {
> - struct pid *tgid;
> -
> bpf_preload_ops = NULL;
> -
> - /* kill UMD in case it's still there due to earlier error */
> - tgid = umd_ops.info.tgid;
> - if (tgid) {
> - kill_pid(tgid, SIGKILL, 1);
> -
> - wait_event(tgid->wait_pidfd, thread_group_exited(tgid));
> - umd_cleanup_helper(&umd_ops.info);
> - }
> - umd_unload_blob(&umd_ops.info);
> + free_links_and_skel();
> }
> -late_initcall(load_umd);
> -module_exit(fini_umd);
> +late_initcall(load);
Since load_skel invokes the loader which resolve kfuncs, and if the module is
builtin, btf_vmlinux's kfunc_set_tab is not fully populated yet (other
register_btf_id_kfunc_set calls may also happen in late_initcall), so I think it
may be a problem.
When I worked on it we didn't have this case where BPF syscall can be invoked at
this point, but obviously that is true now. So I think ordering it later than
late_initcall (or moving register_btf_id_kfunc_set invoking initcalls before) is
required. WDYT?
> [...]
--
Kartikeya
